power in Russia was controlled by the Czars who reacted with a firm hand against any attempt at reform by some social groups. At the out-set of the 20th century, ...
Journal of Educational Psychology 2000, Vol. 92. No. 1,107-116
Copyright 2000 by the American Psychological Association, Inc. 0022-0663/00/55.00 DOI: 10.1037//0022-0663.92.1.107
Two Procedures to Improve Instructional Text: Effects on Memory and Learning Eduardo Vidal-Abarca, Gabriel Martfnez, and Ramiro Gilabert University of Valencia The goal of this study was to compare 2 approaches for improving instructional text. The first was based on the procedure created by B. K. Britton and S. GUlg6z (1991) as derived from W. Kintsch and T. A. van Dijk's theory (1978). It emphasized the reduction of the reader's inferential activity. We created a second method that was inspired by theories of narrative comprehension (P. van den Broek, 1990; A. C. Graesser, M. Singer, & T. Trabasso, 1994). We oriented it toward triggering causal inferences in the reader. Alternative versions of an original passage on history were elaborated for each of the 2 methods. Sixty-four 8th graders read either the original passage or one of the revised versions and were tested on memory (i.e., recall) and learning (i.e., inference questions). Only the 2nd procedure produced benefits on inferential learning, though both procedures had a limited effect on recall.
School textbooks are an important tool for learning; consequently, their quality has long been of concern to professionals in many areas of educational research and practice. One way this concern has been expressed is in research aimed at testing the efficacy of revising the textbooks to improve student learning. One of the most cited studies of text revision was conducted by Britton and Gulgoz (1991). The procedure Britton and Gulgoz (1991) used was based on a computational method devised by Kintsch and van Dijk (1978), as implemented in a computer program (Miller & Kintsch, 1980). Britton and Gulgoz used Miller and Kintsch's computer program to analyze a text about the Vietnam war and identify the text locations at which the reader should make an inference to establish coherence, because there was no overlap between sentences. Britton and Gulgoz then repaired the text at each of those locations by augmenting argument overlap. The following excerpts will clarify the procedure.
the field, pressure to strike directly at North Vietnam had begun to build. Revised passage: By the beginning of 1965, American officials in both South Vietnam and the U.S. had begun to focus on North Vietnam as the source of the continuing war in South Vietnam. The South Vietnamese Army was losing the ground war against North Vietnam and this caused frustration among the American officials. The frustration lead to the bombing of North Vietnam. The original passage used different expressions to refer to the same concept. For example, Saigon is South Vietnam, Hanoi is North Vietnam, and ARVN is the South Vietnamese Army. The repetition of these words in the revised version increased referential coherence and reduced reader inference needs. Britton and Gulgoz (1991) also repaired the text by making relationships between ideas more apparent. For example, the connection among frustration, inability of the ARVN to defeat the enemy, and pressure to strike directly (i.e., bombing attacks) was made explicit in the revised version by specifying who was frustrated (i.e., the American officials) and adding linguistic expressions to make explicit the causal chain among the events.
Original passage: By the fall of 1964, Americans in both Saigon and Washington had begun to focus on Hanoi as the source of the continuing problem in the South. As frustration mounted over the inability of the ARVN to defeat the enemy in
The positive effects of augmenting argument overlap on comprehension can be explained by Kintsch and van Dijk's (1978) theory, which predicts that comprehension processes must be cyclical because of limitations in the working memory. There is a memory buffer in which some propositions from one cycle of processing are retained to form a connection with the propositions of the next. If there is no overlap between the contents of the buffer and the new input, the reader often needs to make inferences to make the textbase coherent. When a text insists that readers make a lot of inferences, many readers will likely fail to make a number of them. Thus, parts of the text will then be disconnected from the whole, and reader recall will decrease. When the argument overlap is augmented, readers need not make as many inferences to link sentences, and reader recall should increase.
Eduardo Vidal-Abarca, Gabriel Martfnez, and Ramiro Gilabert, Department of Developmental and Educational Psychology, University of Valencia, Valencia, Spain. Earlier versions of this article were written by Eduardo VidalAbarca during a stay at the Department of Psychology, University of Georgia. This article was supported by Grant DOGV 2525 from Direcci6n General de Enseiianzas Universitarias e Investigation de la Generalitat Valenciana. We thank Bruce K. Britton, Eileen Kintsch, Herre van Oostendorp, Jean-Francois Rouet, and Art Graesser for their helpful comments on earlier versions of this article. Correspondence concerning this article should be addressed to Eduardo Vidal-Abarca, Departamento de Psicologia Evolutiva y de la Educacion, University of Valencia, P.O. Box 22045, 46080 Valencia, Spain. Electronic mail may be sent to eduardo.vidal-abarca@ uv.es.
The procedure based on argument overlap has two key 107
108
VEDAL-ABARCA, MARTINEZ, AND GILABERT
features. First, breaks in coherence are defined as the absence of argument overlap. Second, lack of coherence is repaired by adding information to eliminate some of the inferences the reader must make. As a consequence, the procedure tends to concentrate the revision activity on local coherence problems because of the nature of the theory and the consequent computational procedure on which it is based. There is some uncertainty as to whether this revision procedure promotes deep levels of comprehension (i.e., the situation model level) rather than comprehension on only a superficial level (i.e., the textbase level). The distinction between these two levels is useful when explaining the differences between text memory and learning from a text (Kintsch, 1998). The textbase level is closer to text memory. The situation model level is closely related to learning from a text (e.g., to make inferences), which requires the integration of text information with the reader's prior knowledge. It might happen that text revision oriented toward increasing argument overlap does not improve the situation model level of understanding because it reduces the need for readers to activate their prior background knowledge. An example will clarify this possibility. We added information in italics to a passage about the Russian Revolutions to repair the lack of coherence defined in terms of argument overlap. During the 19th century Russia had managed to stay on the side-lines of the political, economic and scientific revolutions which had socially transformed the rest of Europe. Political power in Russia was controlled by the Czars who reacted with a firm hand against any attempt at reform by some social groups. At the out-set of the 20th century, the nobility and the clergy were still the dominant social groups. The middle class was practically non-existent and the minority social group of the proletariat pushed for revolution. A number of inferences were needed to fill in the breaks in coherence apparent in the original passage: (a) that it was in Russia where the czars had power, (b) that some of the social groups described in the second paragraph were the protagonists of the attempts at reform, and (c) that the proletariat was a social group. All these inferences were eliminated in the revised version by increasing argument overlap. These changes might improve superficial levels of understanding without increasing deep comprehension because they reduce reader inferential activity necessary to integrate the text information into the reader's prior background knowledge. Our study was aimed at testing this possible limitation of the revision procedure based on argument overlap and testing an alternative method oriented toward increasing the reader's inferential activity. Our alternative procedure is based on the causal constructionist model of narrative text comprehension. It emphasizes the reader's attempts to explain goal and causal structures in the passage. Two theories of narrative text comprehension are relevant to this model: (a) the comprehension model of narrative texts proposed by Trabasso and his colleagues (Trabasso, Secco, & van den Broek, 1984; Trabasso & van den Broek, 1985; van den Broek, 1990) and (b) the constructionist theory of
narrative comprehension formulated by Graesser and his colleagues (Graesser, Singer, & Trabasso, 1994; Zwaan, Langston, & Graesser, 1995). Our new procedure is based on the causal constructionist model. Breaks in coherence in history texts are detected using the causal-time method created by Trabasso et al. (1984) to analyze stories, because understanding historical events implies building a mental temporal-causal model of historical events (Beck, McKeown, & Gromoll, 1989; Perfetti, Britt, Rouet, Georgi, & Mason, 1994). Breaks in coherence occur when the reader needs to make inferences to causally connect two ideas. To repair the text, we added to the text (a) information to trigger the reader's causal antecedents and (b) superordinate goal inferences. An example taken from the Russian Revolution text will clarify this procedure. Numbers in parentheses indicate the order of each idea. Information in italics was added to repair the original text. (1) During the 19th century Russia had managed to stay on the side-lines of the . . . revolutions (2) which had socially transformed the rest of Europe. (3) Political power was controlled by the Czars (4) who reacted with a firm hand against any attempt at reform. (4b) There was a great disparity between rich and poor: (4c) the former controlled the whole of society (4d) while the latter had neither rights nor economic resources. (15) The situation of economic dependence was strongly denounced by Lenin, (16) founder of the Bolshevik Party,... (18) that would later bring about the socialist revolution, (18b) as a means of implementing radical changes.
The added ideas 4b-4c-4d (i.e., control by the rich vs. no rights for the poor) elaborated Ideas 1 (i.e., Russia took no part in revolutions) and 3 (i.e., power was controlled by the Czar), providing readers with a clearer picture of the situation in Russia at that time. This should help readers infer that Ideas 1 and 3 are causal antecedents for the socialist revolution in Russia (Idea 18). The added sentence, 18b (i.e., radical changes), expresses a superordinate goal that helps readers link Ideas 1 and 3 with the announcement of the socialist revolution (Idea 18) and the two revolutions in 1917 described later in the text. It should be noted that argument overlap between the new information and that already present in the text was explicitly avoided. The causal constructionist model of narrative text comprehension can explain why these changes may improve superficial and deep comprehension. According to this model, when a reader reads a story, he or she frequently makes inferences to link the events and states described in the text, thus ordering them into a causal chain (van den Broek, 1990). As the reader encounters an event, he or she attempts to find adequate causal justification for it. If the preceding event does not fulfill the criteria for making a causal connection, then a coherence break occurs. A number of inferences may be required, two of which are critical for establishing causal explanations: causal antecedent and superordinate goal inferences (Graesser et al., 1994). They are highly dependent on the reader's background knowledge. Inferences will likely be made when the reader is reading narrative prose because it is very familiar to the reader, but inferences are less likely to be made when he or
IMPROVING INSTRUCTIONAL TEXT
she is reading expository prose. Therefore, adding information that helps trigger the reader's inferences necessary to connect events causally will result in a coherent mental representation in the reader's mind. Our procedure has clear practical applications to school textbook design because of its features that clearly distinguish it from the argument-overlap procedure. First, breaks in coherence are detected by examining the difficulty readers have in establishing meaningful connections between text ideas. Second, changes are oriented toward triggering reader inferences in place of eliminating them. Our procedure emphasizes the role of constructive activity by the reader and pays attention to local and global coherence of the text examining the causal connection between near and distant events. We tested the efficacy of the two above-mentioned revision procedures by elaborating three revised versions of the Russian Revolution text. In the first text, we repaired the original passage by reducing the reader's inferential activity by increasing argument overlap. We repaired the second text by inserting information to trigger reader causal inferences. The third text version was a combination of the other two. According to the causal constructionist model and Kintsch's (1994, 1998) theory of text comprehension, a number of predictions can be made about the role of the argument overlap and the causal constructionist changes over the superficial and deep levels of comprehension. Argument overlap changes will not produce positive effects at the deep level of understanding; only the causal constructionist changes will be effective at this level, because they will cause readers to make inferences that integrate text ideas into their prior background knowledge. Regarding the superficial level of comprehension, Kintsch's (1994,1998) theory and the constructionist model predict that causal constructionist changes will improve reader recall over the original passage as they contribute to link text ideas in the reader's mind. However, these two theories differ in their predictions regarding the role of argument overlap in regard to recall. Whereas Kintsch's theory emphasizes the role of argument overlap, the causal constructionist model does not. Therefore, according to the former, argument overlap changes would contribute to improving the reader's recall, whereas the latter predicts a marginal effect of these changes.
Method
Participants and Materials Sixty-four eighth graders, with an average age of 13 years, 5 months, participated in the study. The sample consisted of 43% female students and 57% male students, all of whom attended a public school in Valencia, Spain, and came from average-income families. We used a text about the Russian Revolution from an eighthgrade history textbook (Anaya Publishers, 1987). The Spanish version contained 522 words (see Appendix A). The children had yet to study a lesson on the Russian Revolution.
109
Argument-Overlap Version To write this version, we propositionalized the original passage following the procedure developed by Bovair and Kieras (1985), resulting in 180 propositions. We men used a procedure very similar to the one used by Britton and Gulgoz (1991) tofindgaps in the argument overlap between propositions. We could not use Miller and Kintsch's (1980) computer program as Britton and Giilgdz had because of computer language compatibility problems. Padilla and Sanjose (1996) then wrote a program (following Miller and Kintsch's rules) that was designed especially to solve the computer language problems and that produced results identical to those of Miller and Kintsch. The program proceeds in a series of cycles; each sentence represents a cycle. First, the program inputs the first sentence's propositions; then it inputs the second sentence's propositions and looks for argument overlap between the propositions from the two sentences. Coherence is found when the second sentence contains an argument that overlaps with another in the first sentence. Then the program inputs the third sentence, looks for coherence, and continues in this way until the end of the text. When the program cannot find an argument overlapping between propositions, it comes to a halt, because it cannot make inferences. If the lack of connection is not repaired, the program continues with the first unconnected sentence as if it were the first sentence of a new passage. When the original text was run in the program, eight shutdowns occurred. To repair them, we used Rules 1 and 3, implemented by Britton and Gulgoz (1991). Thus, we rewrote the sentences so that there was a repetition between words, and we made information explicit that was previously implicit in the text. We did not use Rule 2 (i.e., change the order of the parts in a sentence to present first the previously known information followed by the new) because no break of coherence within our text needed to be repaired in this manner. As opposed to Britton and Gulgoz, we used Rules 1 and 3 with two constraints: (a) only words already present in the text to produce the argument overlap were added to the text and (b) the original text remained intact; we limited ourselves to adding certain expressions where breaks in coherence were detected. We repaired the eight shutdowns by making explicit the following local relationships (text added is in italics): 1. Five causal relationships (e.g., ". . .soldiers, and students joined the ranks of the strike held by the metallurgy and textile workers of Saint Petersburg. As a consequence of the strike, the power of the czars was replaced by...") 2. Two empty slots within the sentence schema filling temporal and locational information (e.g., "Political power in Russia was controlled by the Czars"). 3. One superordinate-subordinate relationship (e.g., "the minority social group of the proletariat pushed for revolution"). The changes made relationships between sentences explicit, reducing the need for readers to make inferences (see the versions in Appendix A). Thirty-two words were added to this version, which made it 6.1 % longer than the original passage.
Causal Constructionist Version To elaborate this version with our procedure, we used the causal-time method of analysis to detect the lack of coherence (Britt, Rouet, Georgi, & Perfetti, 1994; Trabasso et al., 1984; van den Broek, 1990). The method has three steps. First, the text was divided into idea units or events, each consisting of a subject and a predicate (the original text had 52 idea units). Second, the idea
110
VIDAL-ABARCA, MARTINEZ, AND GILABERT
units were classified into main events and supporting information. Main events are those that are essential to the comprehension of the story. Table 1 shows the list of main events: sociopolitical conditions before the revolutions, the revolutions themselves, their immediate antecedents, events describing actions of the main protagonists (Lenin and Kerenski), and the consequences of each of the revolutions. Supporting information are supporting events (e.g., other revolutions had socially transformed Europe, the agricultural farm labor social group was the largest group) and specific details (e.g., the farm group comprised four-fifths of the population). The third step was to represent all of the relationships between the different idea units in a network. These relationships may be of two types: cause and intersection. The causal relationships must comply with four criteria (van den Broek, 1990). According to the temporal priority criterion, a cause never occurs after a consequence. According to the criterion of operativity, a cause is active when the consequence occurs. The necessity in the circumstance criterion reflects the fact that if the cause had not happened, then the consequence would not have taken place, given the circumstances of the story. The sufficiency in the circumstance criterion indicates that if the cause occurs then the consequence will likely occur as well, given the circumstances of the story. Temporal priority and operativity are required for a causal relation. Necessity and sufficiency can be present in varying
degrees. It is also possible that a certain degree of uncertainty at the moment of establishing a causal relationship between historical events may be practically impossible to eliminate completely in concrete cases (Perfetti et al., 1994). Intersection relationships are those that do not imply a causal relationship but express motives, temporal coincidence, or temporal sequencing. All of the relationships could be either explicit or implicit. By establishing causal connections between events, we detected coherence breaks. These breaks were produced when the reader needed to make inferences to connect causally two events. To repair the text we analyzed the inferences that the reader had to make to connect the main events. We paid special attention to causal antecedents and superordinate goal inferences as they are critical in explaining historical events (Beck, McKeown, Sinatra, & Loxterman, 1991; Perfetti et al., 1994). To help readers, we introduced information to trigger reader inferences. This information was of two types: (a) elaborations of a text idea to clarify its meaning and (b) part of the ideas that readers have to activate to infer the connection between text events. In all cases we avoided using terms already present in the original text so as not to increment the overlapping of arguments in this version. We also avoided making the causal relationship between ideas completely explicit, always leaving room for reader inferences. We presented an example of an elaboration of a text idea earlier in the article. The example referred to the disparity between rich
Table 1
List of Main Events in the Original Text and Ideas to Trigger the Reader's Inferences Added on the CCandAO + CC Versions CC and AO + CC versions
Original passage 1. 3. 10. 11. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 26. 27. 29. 30. 31. 33. 34. 38. 39. 40. 41. 42. 43. 44. 45. Note.
Russia apart from revolutions Power controlled by the czars Economic growth in Russia Foreign capital loans Lenin denounces economic dependence Lenin founder of Bolshevik Party Bolshevik Party formed by revolutionary groups Bolshevik Party brought about socialist revolution Defeat of Russia in the Japanese-Russo War Mobilization of labor and uprisings Revolution of 1905 Goal: to eliminate the czarist regime To make czar follow politically liberal lines Formation of the "Duma" February 1917 revolution Russia was defeated by German army Executive committee of the Duma came to power Kerenski presided over the executive committee of the Duma Soviet Congress came to power October 1917 revolution Kerenski decided to continue the war Bolsheviks forced Kerenski to flee Communist Party became the ruling force Lenin in the government Socialist thesis put into practice State debts nullified Economy nationalized Land redistributed Inhabitants given the status of citizens
Rich control versus no rights of poor Problems not solved; Russian debts Radical changes
People protest about poverty and injustice Situation remained equal
Kerenski against Bolsheviks
CC - causal constructionist; AO = argument overlap.
111
IMPROVING INSTRUCTIONAL TEXT and poor as an elaboration of two text sentences describing the situation in Russia at the end of the 19th century. The following example illustrates the second type of information inserted into the text: (10) Between 1881 and 1914 economic growth took place in Russia (11) thanks to foreign capital loans . . . (14b) The foreign loans not only did not solve the problems of the population but (14c) forced Russia into ever greater debt with even more countries. (15) The situation of economic dependence was strongly denounced by Lenin. The causal connection between Ideas 11 (i.e., foreign loans) and 15 (i.e., economic dependence) requires the reader to formulate a complex chain of inferences: (a) the loans create debts that must be paid to the lender; (b) many loans are effected to generate wealth and thus solve economic problems; (c) if this wealth is not attained, one is left poorer and more dependent on the loaning entity; (d) Russia should have never sought the riches that she had hoped for given that the text claims that Russia developed an economic dependency on other countries. Ideas 14b and 14c are a part of this complex chain of inferences that should trigger the rest of the reader's inferences. This should contribute to the understanding of the causal connection between Ideas 11 and 15. Six sentences containing a total number of nine idea units were added to this version. Some of these ideas connected very distant ideas (e.g., 4b-4c—4d), whereas others connected ideas from the same or consecutive episodes (e.g., 14b-14c, and 34b, respectively). When we ran this version through Padilla and Sanjose's (1996) computer program, the same eight shutdowns found in the original passage also were present in this version. This is the result we were looking for, as we wanted to get the reader to make inferences rather than replace inferential gaps with explicit information. Eighty-eight words were added to this version, which made it 16.9% longer than the original passage. The changes can be seen in Appendix A.
Argument Overlap + Causal Constructionist Version To elaborate this version we simultaneously implemented the two changes described previously. The addition of only two words was required to provide argument overlap between one of the sentences from the causal constructionist version and the original passage so that there was no shutdown while this version was run through Padilla and Sanjose's (1996) computer program. Thus, 122 words were added in this version, which made it 23.4% longer than the original version.
Procedure The study was conducted during two separate sessions. In the first session the children answered a background knowledge test composed of 11 multiple-choice questions. It covered information on the main events stated in the text (e.g., who founded the Bolshevik Party, what were the social and economic conditions of Russia at the beginning of the 20th century, what were the consequences of the different revolutions in Russia, etc.). There were no differences among groups, F(3,63) = 0.292, MSE = 4.119 ( A / = 3 . 8 1 , 5 D = 1 . 7 6 ; M = 4.00, SD = 2.31; M = 4.06, SD = 2.26; and M = 4.12, SD = 1.71, for readers of the original text, argument overlap version, causal constructionist version, and the most complete version, respectively). Immediately after completing the prior-knowledge test, each participant was given one of the four possible versions of the text and asked to study it closely for 18 min. After that time the text was
collected, and the children were asked to write down everything they remembered from the text. The second session took place 1 week later. First, children wrote once again everything they could remember about the text. Then they were handed the text version they had read in the first session. They were asked to answer an inference test composed of 7 open-ended questions and were free to consult the text at any time.
Dependent Measures and Scoring We made two independent measurements: recall and inferences. The former tested memory, and the latter tested learning.
Recall Measures We scored the children's recall protocols against the list of idea units in the original text. These idea units may be either main events or supporting information, according to the criteria explained in the Procedure section. We also scored separately the erroneous information present in the protocols as it was related to the level of learning reached by the students. Erroneous information was defined as either information contradictory to text ideas (e.g., the clergy was the class most dominated) or erroneous inferences that were contradictory to real events (e.g., the Bolshevik Party consisted of the nobility and the clergy). The original passage contained 52 idea units, 29 of which were categorized as main ideas, the rest of which (23) were supporting information.
Inference Test This test consisted of 7 open-ended inference questions. All of them referred to issues regarding causal antecedents, superordinate goals, and consequences of the revolutions (see Appendix B). To answer the questions, children had to make inferences relating separate text ideas. The following example illustrates the kind of questions readers answered: "In the second paragraph the text states that 'the minority proletariat pushed for revolution.' What did they hope to accomplish with the revolution?" To answer this question correctly, the students had to infer that the proletariat was a class of people who had no political or economic power and that because of this they desired a change in the extant situation (i.e., the revolution). The goal of the revolution was to achieve changes necessary for the elimination of the czarist regime. The information of Ideas 4b-4c-4d (disparity between rich and poor), 18b (revolution as a means of implementing radical changes), and 22b (protest about poverty and social injustice) inserted into the causal constructionist version would facilitate those inferences. Each response was scored between 0 and 2 points according to its level of completeness; the maximum score was 14. Specific criteria for each question were elaborated. The recall and inference tests were independently scored by two raters. Interrater correlations were .83 and .92 for recall and inferences, respectively.
Results Analyses We conducted analyses of variance, including two between-subject factors—argument overlap changes (present vs. absent) and causal constructionist changes (present vs. absent)—on two of the measures (i.e., inference test and erroneous information included in the recall protocols). Analysis of proportional recall incorporated an additional
112
VIDAL-ABARCA, MARTINEZ, AND GILABERT
within-subject factor (i.e., level of importance of idea units) with two levels (i.e., main events vs. supporting information). The two sorts of changes were orthogonal. For argument overlap, the argument overlap version and the argument overlap + causal constructionist version included these changes, whereas the original passage and the causal constructionist version did not. Likewise, causal constructionist changes refer to the presence of these changes (the causal constructionist and argument overlap + causal constructionist texts) or their absence (original passage and argument overlap version). To test statistical differences between pairs of text versions, we conducted pairwise / tests.
Text Recall To compare the different text versions, we first analyzed the recall of the idea units that were common to all four texts (i.e., those comprising the original version). The percentage of main ideas and supporting information recalled on each version can be seen in Table 2. In regard to immediate recall, students who read the passages with argument overlap changes present (i.e., argument overlap and argument overlap + causal constructionist versions) obtained higher scores (M = 15.3, SD = 12.1) than students who read the passages without those changes {M = 11.3, SD = 10.1). This difference was not statistically reliable, F ( l , 60) = 3.307, p = .074, MSE - 158.925. A parallel result was obtained for the causal constructionist variable. Students who read passages with these changes present (i.e., the causal constructionist and argument overlap + causal constructionist versions) recalled more idea units (M = 14.9, SD = 11.3) than those who read the versions with those changes absent (M — 11.7, SD = 11.2), This difference was not significant, F(l, 60) = 1.966. The interaction between the argument overlap and causal constructionist variables was not statistically reliable, F(l, 60) - 0.002. Only the passage containing the two sorts of changes (i.e., the argument overlap + causal constructionist
Table 2 Percentages of Ideas for Each Category Recalled Immediately and 1 Week After Reading Each Text Version Text version Original M SD AO M SD CC M SD
AO + CC M SD
Immediate recall Total M S
Delayed recall Total M S
9.4 7.3
6.5 9.9
13.0 10.2
3.2 3.1
2.4 3.0
4.1 5.1
13.2
9.5 8.8
17.9 13.0
6.6 5.0
4.7 5.6
9.0 8.3
13.2
16.6 11.5
9.0 5.6
6.1 4.6
7.3 7.3
4.6 4.0
9.9 7.5
6.4 7.8 4.6 17.1 18.1 15.8 9.0 6.1 7.7 6.2 11.1 15.3 Note. M = main ideas; S = supporting information; AO argument overlap; CC = causal constructionist.
version) was significantly better than the original passage, r(30) = 2 . 3 2 , p < . 0 5 . The main variable of level of importance of idea units did not produce significant differences, F(\, 69) = 0.698; however, an interaction effect between the variables of causal constructionist changes and level of importance of idea units was apparent, F(\, 60) = 17.036, p < .01, MSE = 73.374. Students who read the version with causal constructionist changes present (i.e., causal constructionist and argument overlap + causal constructionist versions) recalled a higher proportion of main events (M = 17.3, SD = 13.4) than of supporting information (M — 12.4, SD = 8.1). We observed the opposite trend for students who read the passages without the additional information (i.e., original and argument overlap version). They obtained lower scores for main events (M = 8.0, SD = 9.4) than for supporting information {M = 15.5, SD = 11.7). The interaction between argument overlap and level of importance of idea units was not significant, F(l, 60) = 1.397; neither was the triple interaction among the three main variables, F(l, 60) = 0.317. Thus, the two kinds of changes improved student overall recall, though this benefit was not statistically reliable. However, the two types of changes had a very different effect on the recall of main versus supporting information. Causal constructionist changes improved reader recall of main events over low important information; argument overlap changes had the opposite effect. It should be noted that causal constructionist changes were aimed at triggering reader inferences to link the main events in the text. Apparently, these changes accomplished their goal. There was a very similar trend for delayed recall. Students who read passages with argument overlap changes obtained better proportional recall (M = 6.52, SD = 7.14) than those who read passages without these changes (M = 4.60, SD = 5.32). Likewise, this difference was not statistically reliable, F(l, 60) = 2.556, p > .10, MSE = 46.311. Causal constructionist changes did not produce significant differences, F(l, 60) = 0.751, either. However, students who read passages with these changes scored higher {M = 6.1, SD = 6.5) than those who read versions without them (M = 5.0, SD = 6.2). The interaction between argument overlap and causal constructionist changes was not statistically reliable, F(l, 60) = 2.015, either. Only the passage with argument overlap changes (i.e., the argument overlap version) was reliably superior to the original passage, r(30) = 2 . 3 6 , p < . 0 5 . The Ievel-of-importance-of-idea-units variable did not produce significant differences, ^ ( 1 , 60) = 0.001. The same interaction effect between causal constructionist changes and level of importance of idea units observed for the immediate recall was apparent here, F ( l , 60) = 9.295, p < .01, MSE = 29.851. Students who read the version with causal constructionist changes recalled a higher proportion of main events (M = 7.54, SD = 7.40) than of supporting information (M = 4.62, SD = 5.17), whereas the opposite trend was true for students who read the version without these changes (M - 3.56, SD = 4.59; M = 6.52, SD = 7.24, for main events and supporting information, respectively).
113
rMPROVING INSTRUCTIONAL TEXT The interaction between argument overlap and level of importance of idea units was not significant, F ( I , 60) = 1.397; neither was the triple interaction among the three main factors, F ( l , 60) = 0.317. Thus, neither argument overlap nor causal constructionist changes had a clear effect on overall recall. However, the argument overlap version was significantly better than the original passage with respect to delayed recall. Clearly, we can conclude that the two methods of revision produce different effects on recall: Whereas causal constructionist changes strengthened the recall of main events over supporting information, the opposite was true for the argument overlap changes. An explanation for this is that the first sort of change triggers causal antecedent and superordinate goal inferences in the reader's mind. To support this idea with additional evidence, we computed two Pearson's correlation coefficients: (a) recall of inference-triggering information and memory of main events, and (b) recall of inference-triggering information and recall of supporting information—one pair of coefficients for the causal constructionist version and one pair for the argument overlap + causal constructionist version. As can be seen in Table 3, for the students who read the causal constructionist version, the correlation between the recall of the inferencetriggering information and the recall of main events is significant, whereas the correlation between the recall of the inference-triggering information and that of supporting information is null for both immediate and delayed recall. The correlation coefficients for the version with both sorts of changes together are different. Neither of the two coefficients is statistically reliable for immediate recall, but they are reliable for delayed recall. This pattern of results is consistent with the conclusion that the inference-triggering information inserted in the passages actually played the hypothesized role (i.e., helping readers establish causal links among main text events). We also analyzed the amount of erroneous information present in the recall protocols as it is related to the level of learning reached by the students (see Table 4). In the immediate-recall condition, causal constructionist changes produced significant differences, F ( l , 60) = 6.629, p < .05, whereas argument overlap did not, F ( l , 60) = 0.135. Students who read the passages with the information inserted into the text to trigger reader inferences included fewer erroneous ideas in their protocols of recall (M = 2.56,
Table 3 Correlations Between Recall of Inference-Triggering Information and Either Main Ideas (M) or Supporting Ideas (S)
Delayed
Text version
M
S
M
S
CC AO + CC
.56* .21
.01 -.18
.73** .61*
.04 .47*
Note,
CC = causal constructionist; AO - argument overlap.
*p